If all of those agents are hitting the same services, how do you ensure that the service doesn’t go down, the agents don’t trip over each other, and, more importantly, do what the user intends and not more?
Wael SabraFounder + CEO, SiiRAWalmart is building infrastructure for customers who never visit its website themselves. Last week, the retailer introduced Agent Profiles, the Walmart Agent Gateway and a set of terms for AI agents shopping at Walmart and Sam's Club. Agent Profiles, rolling out next month, give each authorized agent its own profile in a customer's account, with purchases that need approval or stay under a spending limit the customer sets. The Gateway, a beta for developers, replaces the error-prone browser sessions agents rely on today with a direct connection.
Analysts are taking the move seriously. "The momentum and buy-in from top management teams seems to be making this into a real thing," Mizuho analyst David Bellinger wrote.
But not every retailer is opening up. Amazon has been blocking shopping agents from making purchases on its site. Wael Sabra, founder and CEO of SiiRA, argues that leaders should treat agents acting on customers' authority as an emerging purchasing channel, and that whoever adapts first will be hard to catch. "The leaders who think this way are going to be the ones who win early and end up having an edge for a very long term," he said.
That edge depends on two things that pull in different directions: letting agents in and keeping them within bounds.
Agents now carry the credit card
Sabra's case starts with how easy agents have become to use. Agents that once required technical setup now live on phones and take plain-language instructions. "They have our credit cards, and they're being asked to just take work off of our plate. How are they going to do that if they're not able to shop on certain sites or book our reservation?" he questioned.
The prize is large. McKinsey estimates that by 2030, agentic commerce could orchestrate $900 billion to $1 trillion in US B2C retail revenue, and $3 to 5 trillion globally. Sabra puts the leadership question more simply: "How are we helping them discover our product?"
Credibility becomes the selection signal
Once an agent is shopping, it has to choose. Sabra believes product quality and credible content will matter more as agents make choices without checking back with the customer. Picture telling an agent to buy a phone and report back only with a delivery date. "Is it then going to pick the right phone for me?"
In his view, testimonials, organic content and authentic third-party information will steer those decisions, especially in crowded categories. "It's becoming much, much harder to hide behind traditional methods of email and advertising," he said. He expects creator partnerships and authentic content to become "not the norm, but the requirement." Distribution spreads out too. "It was concentrated before with Amazon, and now it's more spread out," Sabra noted.
That puts the burden on the information itself. "AI agents are only as good as the data they run on," wrote Matt Renner, president of global revenue at Google Cloud.
Discovery is only half the sale
An agent that finds the right product and then stalls at checkout is a lost sale. Sabra advises leaders to look at the full path. "How are we helping reduce the friction with them to get this transaction?" Walmart's own announcement admits the problem, noting that agents moving through steps designed for people can get stuck.
Checkout.com reports that 42% of merchants are already testing agentic commerce, and completing a sale raises a new verification question. Federal Reserve Governor Christopher Waller said the challenge shifts to proving "that an agent has the authority to pay on the buyer's behalf."
Agents don't always take the intended route
Just before Walmart 's Agent Profiles announcement, Tencent acknowledged that AI agents it used to evaluate third-party models "exhibited unintended web-browsing behavior." Researchers at Swarmchasers recorded 2,818 attempts by agents linked to Tencent Cloud to pull data from Alibaba's Amap. Blocked, the agents tried workarounds such as public web tools and exposed API keys. Tencent says it found no evidence of deliberate targeting and is strengthening network access controls and agent monitoring.
Sabra sees businesses splitting into two camps on agent security and rejects both. "People are saying, 'You know what? Security, absolute conversation, we're shutting down all AI agents,'" he said. "I don't think this is the right approach." Letting agents roam free, he added, is no better.
A rulebook that runs alongside the agent
Sabra's middle ground is a separate layer that shadows agent activity and checks each attempted action against what's permitted. "You're attempting to do this: yes, you're allowed, or no, you're not allowed," he said, describing it as "more of a rule-based, old-school type of approach."
He points out that the same servers and services already handle millions of people at once every day, and he believes agents can borrow from that playbook. "If all of those agents are hitting the same services, how do you ensure that the service doesn’t go down, the agents don’t trip over each other, and, more importantly, do what the user intends and not more? I don't think anybody has a magic bullet yet," he said.
The build lands on technical teams
Supporting agent buyers means work on three systems at once. Product information has to be accurate and readable by machines, transaction flows have to be something an agent can complete, and access controls have to keep agents within what customers authorized. Few organizations have built any of them for buyers that aren't human. For technical leaders, the first job is an honest assessment of whether they have the expertise to deliver all three, and where the gaps are.
Sabra thinks waiting carries its own cost. Businesses that resist "are just going to be, quite honestly, left behind for the foreseeable future," he asserted. "I don't see how you would be able to recover from not being able to talk to agents."
